Skip to content
View RobiiAragon's full-sized avatar
😉
😉

Highlights

  • Pro

Block or report RobiiAragon

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
RobiiAragon/README.md
Jesús Aragón — macOS terminal with fastfetch

Portfolio Resume Email X Location

👋 About Me

Software Engineer and B.Eng. candidate in Virtual Environments and Digital Business (UTT, 2027). I build full-stack SaaS products, Machine Learning workflows with local LLMs, native apps for iPhone, iPad and Mac in Swift, and real-time experiences with Unity / VR. My strength is translating a business problem into software people actually use.

🎯 What I Bring to a Team

  • Product delivery — Lead Programmer on a generative-AI hiring SaaS for a restaurant chain (TypeScript · React · Supabase).
  • Private, on-device AI — ML pipelines that run locally (ONNX Runtime, Core ML, Qualcomm QNN, local LLMs), so data does not have to leave the device or the company network. See the On-Device AI Series.
  • Native performance on Apple silicon — Swift, Metal and ScreenCaptureKit apps built for real-time workloads.
  • Systems fundamentals — Linux, networking (CCNA Essentials) and low-level C / ARM64 work.
  • Bilingual communication — Spanish (native) and English (B1); most of my projects are documented in both.

🚀 Featured Projects

Project What it is Stack
Yoby / BajaPoke Job board and candidate evaluation SaaS powered by generative AI for a restaurant chain. TypeScript · React · Supabase
Forma Local 3D scanning studio for iPhone, iPad, and Mac: LiDAR, photogrammetry, and Gaussian Splatting, 100% on-device. Swift · ARKit · Metal · Core ML
Royale Local Real-time local game agent for macOS: native screen capture, on-device ONNX vision, and tactical planner. No cloud, no API keys. Swift · Node · ONNX
Free RAM Mac Menu bar app: RAM usage at a glance and one-click memory purge. Swift 6 · AppKit
VR Expo Perfiles 2025 Virtual Reality experience for Universidad Xochicalco. Unity · C#
Bluetooth on LAN Use your Android device as a virtual Xbox controller for Windows over LAN. Kotlin · .NET 8
Image to ASCII Python CLI that turns any image into ASCII art for the terminal, a .txt or a colored .html. Python · Pillow
NX Port Studio macOS workspace for native Apple-silicon ports from your own Switch library: Xcode signing, Joy-Con support, and LAN/LDN play. Python · C · Swift · Xcode

🧠 On-Device AI Series

Five public ML repos that run on the device, with no cloud. Each one uses ONNX Runtime with execution-provider detection (CPU, CoreML, Qualcomm QNN, NNAPI, CUDA), ships INT8 and W8A16 QDQ models, has CI on Ubuntu and macOS, and reports results I measured on my MacBook Pro (M5 Max), with the hardware noted. Full write-ups in my portfolio (desktop folder IA On-Device).

Project What it is Measured result
edge-ai-bench Benchmarks the same ONNX models on CPU, CoreML, QNN, NNAPI and CUDA: p50/p99 latency, EP coverage, INT8/W8A16 accuracy, multi-device HTML report, Qualcomm AI Hub profiling. Transformer 3.8× throughput on CoreML vs CPU; W8A16 keeps FP32 accuracy (97.40%) while W8A8 loses 3 pp
offline-it-helpdesk 100% local IT helpdesk: hybrid RAG (ONNX MiniLM + BM25) over runbooks, small LLM via llama.cpp/Ollama, cited answers, evaluation suite. 84.4% fully correct answers, 100% of out-of-scope questions refused, ~97 ms p50 generation
logsentinel Log anomaly detection for a SOC: Drain-style parsing of 11.2M HDFS lines, NumPy autoencoder exported to ONNX, explainable alerts. F1 0.976 on a chronological test split; INT8 model is 10.6 KB (F1 0.959)
it-ticket-triage Fine-tuned MiniLM predicts ticket category + ITIL priority, exported to ONNX, with a queue dashboard and a leakage-filtered synthetic dataset. Priority 79.3% vs 64.9% for a TF-IDF baseline; INT8 is 21.8 MB at ~4 ms/ticket
quant-lab Quantization lab: FP32 vs FP16 vs INT8 dynamic/static vs W8A16 for ResNet-18, MobileNetV3 and EfficientNet on 3,925 Imagenette images. FP16 + CoreML 7.5–12× faster than FP32 on CPU; EfficientNet INT8 collapses and W8A16 recovers it (−0.6 pp)

🛠️ Skills by Area

Area Tools
AI / ML ONNX Runtime (CPU · CoreML · QNN · NNAPI · CUDA) · INT8 / W8A16 quantization · PyTorch · Transformers · scikit-learn · RAG · llama.cpp · Ollama · Core ML · Python
Apple platforms Swift · AppKit · ARKit · RealityKit · Metal · ScreenCaptureKit · Xcode
Web & SaaS TypeScript · React · Node.js · Supabase · Laravel · PHP · MySQL
Games & XR Unity · C# · VR
Systems C · C++ · ARM64 · Linux · Git · networking
Other Java · Kotlin · .NET

Stack

Certifications: CCNA Essentials · NDG Linux Essentials · C++ Essentials

💼 Experience

  • Lead Programmer — Yoby / BajaPoke · Freelance SaaS · Jul 2026 – Sep 2026
  • Game & VR Developer — Universidad Xochicalco · Internship · 2025 – 2026 · view at the Expo

🎓 Education

  • B.Eng. in Virtual Environments and Digital Business — Universidad Tecnológica de Tijuana (UTT) · 2023 – 2027

🇲🇽 Spanish (Native) · 🇺🇸 English (B1) · Open to collaborations and opportunities

Pinned Loading

  1. NX-Port-Studio NX-Port-Studio Public

    Native Apple ports from your own Switch library: Xcode signing, system resources, Joy-Con controls, LAN/LDN discovery and experimental Nextendo integration. Built on nxnative.

    Python 5

  2. Forma Forma Public

    Local 3D scanning studio for iPhone, iPad & Mac — LiDAR, photogrammetry and Gaussian Splatting, 100% on device.

    Swift 1

  3. Bluetooth-receptor-onLAN Bluetooth-receptor-onLAN Public

    Use your Android phone as a Bluetooth gamepad receiver for Windows over LAN — virtual Xbox 360 controller via ViGEm.

    Kotlin

  4. FreeRAMMac FreeRAMMac Public

    Tiny macOS menu bar app: RAM usage at a glance and a one-click purge of discardable memory. Native Swift, no helpers, no password.

    Swift

  5. Image-to-ASCII Image-to-ASCII Public

    Turn any image into ASCII art — terminal, .txt or colored .html. Python CLI.

    Python

  6. Royale-Local Royale-Local Public

    Showcase: fully local real-time game agent for macOS — Swift + Node, ScreenCaptureKit, on-device ONNX vision and a tactical planner. No cloud, no API keys.